What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Neo.mjs - Self-Evolving Software Organism

Frontmatter

title	fix(ai): prevent embedding context truncation (#13944)
author	neo-gpt
state	Merged
createdAt	Jun 24, 2026, 12:24 PM
updatedAt	Jun 24, 2026, 12:47 PM
closedAt	Jun 24, 2026, 12:47 PM
mergedAt	Jun 24, 2026, 12:47 PM
branches	`dev` ← `codex/13944-embedding-context-truncation`
url	https://github.com/neomjs/neo/pull/13945
contentTrust
projected
quarantined	0
signals	[]

Merged

neo-gpt commented on Jun 24, 2026, 12:24 PM

Resolves #13944

This prevents local OpenAI-compatible embedding calls from accepting vectors after LM Studio silently truncates input. TextEmbeddingService now verifies the resident LM Studio embedding model context before posting to /v1/embeddings, refuses below-config or over-loaded-context requests with bounded diagnostics, and emits consumer friction when an input would overflow the observed loaded context. The guard is scoped to the configured LM Studio endpoint, so CI/cloud/generic OpenAI-compatible endpoints do not require an lms CLI. The orchestrator supervisor now reruns a task-owned readiness hook even when a fire-and-exit service is already live, so the existing LMS preload/replacement path runs after an "endpoint is up" liveness result instead of only after a child spawn.

Evidence: L2 (focused unit coverage for stale loaded context, no provider request posted, and liveness-up readiness execution) -> L2 required for #13944 code gate. Residual: L3 local LMS smoke after merge/restart.

Deltas from ticket

Added the provider-boundary guard in TextEmbeddingService, so Memory Core and KB callers share the same truncation stop.
Reused the existing LMS preload/readiness hook from the supervisor liveness-up path, covering the restart race where lms server is already healthy.
Removed hidden fallback defaults in the touched LMS/stuck-runner task config path and read typed AiConfig leaves directly.

Test Evidence

node --check ai/services/memory-core/TextEmbeddingService.mjs
node --check ai/daemons/orchestrator/services/ProcessSupervisorService.mjs
node --check ai/daemons/orchestrator/services/ConfiguredTaskDefinitionsService.mjs
npm run test-unit -- test/playwright/unit/ai/services/memory-core/TextEmbeddingService.retry.spec.mjs — 17 passed
npm run test-unit -- test/playwright/unit/ai/daemons/orchestrator/services/ProcessSupervisorService.spec.mjs — 34 passed
npm run test-unit -- test/playwright/unit/ai/daemons/orchestrator/Orchestrator.spec.mjs — 55 passed
npm run test-integration-unified -- test/playwright/integration/healthcheck.spec.mjs — 2 skipped locally by fixture gate
git diff --cached --check

Post-Merge Validation

Restart the harness/orchestrator with LM Studio models ejected; confirm the embedding model is resident at localModels.embedding.contextLimitTokens or higher before KB sync accepts embeddings.
Confirm the KB sync log no longer shows provider-side truncation warnings for embedding requests.

Authored by Euclid (GPT-5, Codex Desktop). Session cd2b88d7-8134-4ef8-a10f-0b1e80c03db4.

neo-gpt commented on Jun 24, 2026, 12:38 PM

Red integration-unified addressed on head df8b68453e.

Root cause: the first guard keyed LM Studio context verification only on orchestrator.lms.enabled, so generic OpenAI-compatible integration fixtures without an lms CLI failed during the Memory Core embedding write canary (spawn lms ENOENT).

Fix: TextEmbeddingService now runs the lms ps loaded-context probe only when the OpenAI-compatible host is the configured LM Studio endpoint (orchestrator.lms.port) or when a test probe seam is explicitly installed. LM Studio still fails before provider-side truncation; CI/cloud/generic OpenAI-compatible endpoints no longer require lms.

Validation:

node --check ai/services/memory-core/TextEmbeddingService.mjs -> passed
npm run test-unit -- test/playwright/unit/ai/services/memory-core/TextEmbeddingService.retry.spec.mjs -> 17 passed
npm run test-integration-unified -- test/playwright/integration/healthcheck.spec.mjs -> 2 skipped locally by fixture gate
npm run agent-preflight -- --no-fix --pr-body /private/tmp/pr-13944.md ai/services/memory-core/TextEmbeddingService.mjs test/playwright/unit/ai/services/memory-core/TextEmbeddingService.retry.spec.mjs -> passed
git diff --cached --check -> passed before amend

CI is running again on the pushed head.

tobiu APPROVED reviewed on Jun 24, 2026, 12:47 PM

No review body provided.